Motion Segmentation and Recognition for Imitation Learning and Influence of Bias for Learning Walking Motion of Humanoid Robot Based on Human Demonstrated Motion

抄録

Two main issues arise in practical imitation learning by humanoid robots observing human behavior – the first is segmenting and recognizing motion demonstrated naturally by a human beings and the second is utilizing the demonstrated motion for imitation learning. Specifically, the first involves motion segmentation and recognition based on the humanoid robot motion repertoire for imitation learning and the second introduces learning bias based on demonstrated motion in the humanoid robot’s imitation learning to walk. We show the validity of our motion segmentation and recognition in a practical way and report the results of our investigation in the influence of learning bias in humanoid robot simulations.